+1 On Nov 19, 2009, at 1:15 PM, Sean Owen wrote:
> I agree that we should have some serious SVM support. I also just > closed out a very old issue on this topic! > > I am all for it if you can submit some code, that you are basically > willing to support. A Hadoop-based implementation is ideal, but > anything consistent is good in my book. > > Post a patch if you'd like to proceed, IMHO. > > On Thu, Nov 19, 2009 at 6:10 PM, Ted Dunning <[email protected]> wrote: >> Mike, >> >> Good to hear from you. >> >> A solid Java implementation of SVM would be of interest. A highly scalable >> implementation would be even more interesting. >> >> On Thu, Nov 19, 2009 at 9:15 AM, mike bowles <[email protected]> wrote: >> >>> I’m a newbie here and am a little timid about raising my hand, but I do >>> want to be helpful, so here goes. If I’m making some mistake in etiquette, >>> please let me know (but be gentle). >>> >>> >>> >>> The most commonly used algorithm for SVM is based on work done be John >>> Platt at Microsoft Reasearch (used for example in libsvm). I’ve attached >>> Platt’s original paper. It’s also still available from Microsoft Research. >>> The commonly used version of the algorithm is based on improvements >>> suggested by Keerti et al. That paper is also attached. The improved >>> algorithm boils down to very few steps. I’ve attached a single page pdf >>> that succinctly summarizes the improved algorithm. >>> >>> >>> >>> I’ve coded this algorithm in C++ and Python and will happily supply code or >>> can easily re-code into Java in a day or so. I hope this is helpful. Do I >>> need to ask that you feel free to comment? >>> >>> mike >>> >> >> >> >> -- >> Ted Dunning, CTO >> DeepDyve >> -------------------------- Grant Ingersoll http://www.lucidimagination.com/ Search the Lucene ecosystem (Lucene/Solr/Nutch/Mahout/Tika/Droids) using Solr/Lucene: http://www.lucidimagination.com/search
